The Bill Was Already Tired
At the post office, the dangerous trays were the ones that arrived looking finished.
A tray of letters thrown together by a drunk could be dealt with. You saw the trouble. You sorted it again. You swore at it. You found the envelopes for Ohio hiding beneath the envelopes for Oregon. A mess announced itself honestly.
But a tray that came down from some bright office already banded, labeled, and stamped—there was a thing to fear. Somebody had made decisions before it got to you. Somebody had decided the work was done. If the bundles were wrong, you were not being asked to sort mail anymore. You were being asked to discover which confident stranger had made a mistake while the clock was running.
That is a special kind of work. It has no glory in it. You are not producing anything. You are standing between a bad decision and the people it will land on.
Now Congress has found a way to make more of that work.
Staffers and outside groups are using the cheerful writing machines to draft legislation. Then the bills arrive at the House Office of Legislative Counsel, where the lawyers are supposed to turn political wishes into words that do not accidentally blow holes in the country.
The office says fixing the machine-written material can take longer than writing the bill from scratch.
Of course it can.
A bill is not a high-school essay with a flag pin on it. You do not get to put a conclusion at the end, add three citations of questionable ancestry, and hope the teacher is hungover. A bill carries a knife in every definition. Change one word and a family in Washington, a tribal government, a small business in Nebraska, or a man who has never heard of the bill can find himself included, excluded, taxed, fined, licensed, erased, or invited to fill out a form until he dies.
The machine is very good at making a sentence look as though it has been somewhere respectable.
That is not the same as knowing where the sentence leads.
One lawyer described the trouble plainly: the box may not understand the difference between a tax credit, a tax deduction, a tax exclusion, and a grant. To the machine, these are neighboring words in a clean little language neighborhood. To a person trying to keep a business alive or pay for medicine, they may be four different doors, three of them locked.
Definitions are worse. Say “state” when you mean the fifty states and suddenly Washington, D.C., or tribal nations may be standing outside in the rain, looking at a program that was supposed to include them. A wrong statutory reference can plant thousands of bugs in the legal system. Not computer bugs, either. The old kind. The kind that acquire offices, forms, enforcement powers, and men who tell you there is nothing they can do because the rule is the rule.
This is what the salesmen never put on the slide.
They show a person asking the machine to draft something. The cursor glides. The paragraphs bloom. Everyone gets twenty minutes back to have a meaningful lunch or stare at another screen. They do not show the tired lawyer with forty years of legal debris in his head reading a polished page one word at a time because somebody, somewhere, trusted a machine that learned the shape of certainty without ever having to live under it.
The machine writes the first draft. The human gets the old job plus a new one: reading for the particular madness of a thing that is fluent and wrong.
It is like hiring a clerk who types two hundred words a minute, never sleeps, and regularly puts the address for a funeral home on a shipment of live chickens. You do not call him efficient because he filled out the labels quickly. You fire him before the chickens arrive.
But we have decided that the correct response to machine speed is to increase the burden on whoever is still capable of embarrassment.
The Legislative Counsel office was not sitting around waiting for a new source of beautiful trouble. It prepared more than thirty thousand bills in the last Congress and more than twenty-one thousand amendments. The work has been growing. The lawyers are, on average, less experienced than they were. More than half have less than a decade on the job.
So naturally the country is sending them a tidal wave of text that looks ready to use and is not.
There is an old office trick in this. When management wants to cut a job, it first makes the job impossible. It adds reports, checklists, new software, meetings about the new software, and a man in a tie explaining that all this will free everyone to focus on the real work. Then, when the workers begin drowning, management looks concerned and introduces a tool to help with the drowning.
The tool creates more water.
The important people will say nobody is proposing that a chatbot enact laws alone. Fine. Nobody says the clerk alone denied your claim, either. The computer flagged it. The supervisor approved it. The department followed procedure. The person who loses has a lovely row of invisible hands to thank.
A machine-drafted bill does not have to be passed whole to do damage. It only has to make the people who check it tired enough to miss one small poison tooth. One sentence. One cross-reference. One definition with a door left unlocked for the wrong fellow and locked for everybody else.
That is the human angle, if we still permit one. Not whether a bot can produce legislative prose. A drunk with a dictionary can produce legislative prose. The question is who pays for the false confidence when the prose arrives at the one place where confidence has to be earned.
There are narrow uses for machines here. The House already has a tool that helps show how proposed changes would alter the U.S. Code. It reportedly returns an error when it cannot place a change with confidence. That is a machine behaving like a decent coworker: it knows when to stop talking.
I respect that more than all the grand promises of artificial intelligence combined.
A tool that says, I do not know, can save a person hours.
A tool that says, Here is a complete bill, when it means Here is a stack of language-shaped rubble, sends the hours downstream to somebody with a bad back, a deadline, and a responsibility no algorithm will ever feel in its metal sleep.
Someday a citizen will stand at a counter with a letter he does not understand. It will contain a rule with a clean number and a cold answer. The person behind the counter will be sorry. The supervisor will point to the statute. The statute will have no face, no memory, and no explanation for why one small word was put there.
Somewhere far behind that counter, a machine will have saved somebody time.
What a rotten thing to save.